-
Regular algorithms: Read all the input and solve the problem.
-
Property testing: Solve the problem by only reading a small part of the input.
-
The concept of ε-farness:
- ε: A constant threshold.
- Is it possible to determine with a certain probability whether a given property is likely to hold or not (ε-far from the property)? Is this what property testing does?
-
Why do property testing?
- When data becomes huge, even polynomial time algorithms can be slow.
- With property testing, it is possible to solve the problem in constant time or even linear time.
- Even if the problem is NP-hard, there is a possibility that property testing can be done.
- It can be used for pruning before solving the problem exactly (rejecting possibilities that are not feasible).
- When data becomes huge, even polynomial time algorithms can be slow.
-
Connections with other fields:
- Approximation algorithms:
- Approximation algorithms usually provide guarantees based on multiplication, such as “$0.878n”.
- Property testing has a slightly different way of providing guarantees.
- However, the techniques of property testing can be applied in some cases.
- Distributed algorithms:
- Distributed algorithms determine the output by randomly looking at a part of a Graph or other structures.
- The approach is similar to property testing.
- Approximation algorithms:
-
Advantages of property testing:
- The algorithms are simple, and the more difficult part is the analysis.
- Experts can handle that.
- It reveals the relationship between global/macro structures and local/micro structures.
- The algorithms are simple, and the more difficult part is the analysis.
-
Property testing is studied for various subjects:
- Monotonicity Property Testing
- Testing Properties of Graphs
- Goal: Obtain necessary and sufficient conditions for properties that can be solved in constant time.
- Some research has succeeded (impressive).
- However, since it is not possible to read all inputs in the first place, information-theoretic methods are used.
-
(Master of Information Science)